S. L. Sondhi et al- Continuous Quantum Phase Transitions

8/3/2019 S. L. Sondhi et al- Continuous Quantum Phase Transitions

1/62

arXiv:cond-mat/9609279v1

29Sep1996

Continuous Quantum Phase Transitions

S. L. Sondhi

Department of Physics, Princeton University, Princeton NJ 08544

S. M. Girvin

Department of Physics, Indiana University, Bloomington IN 47405

J. P. Carini

Department of Physics, Indiana University, Bloomington IN 47405

D. Shahar

Department of Electrical Engineering, Princeton University, Princeton NJ 08544

February 1, 2008

1
http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1http://arxiv.org/abs/cond-mat/9609279v1


2/62

Abstract

A quantum system can undergo a continuous phase transition at the absolute zero

of temperature as some parameter entering its Hamiltonian is varied. These transi-

tions are particularly interesting for, in contrast to their classical finite temperature

counterparts, their dynamic and static critical behaviors are intimately intertwined.

We show that considerable insight is gained by considering the path integral descrip-

tion of the quantum statistical mechanics of such systems, which takes the form of

the classical statistical mechanics of a system in which time appears as an extra

dimension. In particular, this allows the deduction of scaling forms for the finite

temperature behavior, which turns out to be described by the theory of finite size

scaling. It also leads naturally to the notion of a temperature-dependent dephas-

ing length that governs the crossover between quantum and classical fluctuations.

We illustrate these ideas using Josephson junction arrays and with a set of recent

experiments on phase transitions in systems exhibiting the quantum Hall effect.

2


3/62

CONTENTS

I. Quantum Statistical Mechanics: Generalities 6A. Partition Functions and Path Integrals 7B. Example: 1D Josephson Junction Arrays 9C. Quantum-Classical Analogies 12D. Dynamics and Thermodynamics 15

II. Quantum Phase Transitions 16A. T = 0: Dynamic Scaling 16B. T = 0: Finite Size Scaling 20C. The Quantum-Classical Crossover and the Dephasing Length 23

III. Experiments: QPTs in Quantum Hall Systems 25A. Temperature and Frequency Scaling 30B. Current Scaling 32C. Universal Resistivities 35D. Unresolved Issues 36

IV. Concluding Remarks, Other Systems 37ACKNOWLEDGMENTS 38Appendix A 40References 41

3


4/62

A century subsequent to Andrewss discovery of critical opalescence1 in carbon dioxide,

continuous phase transitions continue to be a subject of great interest to physicists. The

appeal of the subject is twofold. First, the list of systems that exhibit interesting phase

transitions continues to expand; it now includes the Universe itself! Second, the formal

theory of equilibrium phase transitions has found applications in problems as diverse as

constructing field and string theories of elementary particles, the transition to chaos in

dynamical systems, and the long time behavior of systems out of equilibrium.

Our purpose in this Colloquium is to give a brief and qualitative account of some basic

features of a species of phase transitions,2 termed Quantum Phase Transitions (QPTs),

that have attracted much interest in recent years. These transitions take place at the

absolute zero of temperature, where crossing the phase boundary means that the quantum

ground state of the system changes in some fundamental way. This is accomplished by

changing not the temperature, but some parameter in the Hamiltonian of the system. This

parameter might be the charging energy in Josephson junction arrays (which controls their

superconductor-insulator transition), the magnetic field in a quantum Hall sample (which

controls the transition between quantized Hall plateaus), doping in the parent compound

of a high Tc superconductor (which destroys the antiferromagnetic spin order), or disorder

in a conductor near its metal-insulator transition (which determines the conductivity at

zero temperature). These and other QPTs raise new and fascinating issues for theory and

experiment, most notably the inescapable necessity of taking quantum effects into account.

Exactly what quantum effects are at issue is a bit subtle. As a corollary of our definition,

all finite temperature3 transitions are to be considered classical, even in highly quantum

1Opalescence is the strong reflection of light by a system (such as an opal) due to fluctuations

in its index of refraction on length scales comparable to the wavelengths of visible light. A liquid

vapor system near its critical point has large density fluctuations on length scales which can reach

microns. This causes the system, which is normally transparent, to have a cloudy appearance.2Henceforth we shall use phase transitions as a shorthand for continuous phase transitions.3In an almost standard abuse of language, we refer to non-zero temperatures as finite.

4


5/62

mechanical systems like superfluid helium or superconductors. It is not that quantum me-

chanics is unimportant in these cases, for in its absence there would not be an ordered state,

i.e. the superfluid or superconductor. Nevertheless, sufficiently close to the critical point

quantum fluctuations are important at the microscopic scale, but not at the longer length

scales that control the critical behavior; in the jargon of statistical mechanics, quantum

mechanics is needed for the existence of an order parameter4 but it is classical thermal

fluctuations that govern it at long wavelengths. For instance, near the superfluid lambda

transition in 4He, the order parameter is a complex-valued field which is related to the

underlying condensate wave function. However, its critical fluctuations can be captured

exactly by doing classical statistical mechanics with an effective Hamiltonian for the order

parameter field (for instance the phenomenological Landau-Ginsburg free energy functional

(Goldenfeld, 1992)).

The physics behind the classical nature of finite temperature transitions is the following:

Phase transitions are quite generally accompanied by a divergent correlation length and cor-

relation time, i.e. the order parameter (e.g. the magnetization in a ferromagnet) fluctuates

coherently over increasing distances and ever more slowly. The latter implies that there is

a frequency associated with the critical fluctuations that vanishes at the transition. A

quantum system behaves classically if the temperature exceeds all frequencies of interest,

and since h kBTc close to the transition, the critical fluctuations will behave classically.This argument also shows that the case of QPTs where Tc = 0, is qualitatively different

and that there the critical fluctuations must be treated quantum mechanically. In the

following we will describe the language and physical pictures that enable such a treatment

4An order parameter is a quantity which is zero in the disordered phase and non-zero in the

ordered state. In systems that spontaneously break some symmetry in the ordered state, the

nature (and value) of the order parameter reflects this broken symmetry. Thus for example in an

Ising ferromagnet, the magnetization is a positive or negative real number indicating the difference

in populations of the up and down spins. See Goldenfeld (1992).

5


6/62

and which have come into common usage among practitioners in the field in the last few

years. Although much of this wisdom, which has its roots in work on quantum Ising models

(Young, 1975; Suzuki, 1976), dates back to the work of Hertz (1976)5, it remains unknown

or poorly understood in the wider community and extracting it from the literature remains a

daunting task. It is our hope here to communicate this set of ideas to a wider audience with

the particular desire to be helpful to newcomers to this field, experimentalists and theorists

alike.

Our discussion is organized as follows. In Section I we introduce the statistical me-

chanics of quantum systems and the path integral (Feynman, 1972) approach to it, which

is an extremely useful source of intuition in these problems. A running theme throughout

this discussion is the intertwining of dynamics and thermodynamics in quantum statistical

mechanics. In Section I.A we describe the general features of a QPT at T = 0 and how a

non-zero temperature alters the physics. This leads naturally to a discussion of what kind

of scaling behavior in experiments is evidence of an underlying QPT in Section I.B. We il-

lustrate this using particular examples from phase transitions in quantum Hall systems. We

end in Section IV with a brief summary and pointers to work on QPTs in other interesting

systems. Readers interested in a highly informative discussion at a higher technical level

should consult the recent beautiful review article by Sachdev (1996).

I. QUANTUM STATISTICAL MECHANICS: GENERALITIES

Before we discuss what happens in the vicinity of a QPT, let us recall some very general

features of the statistical mechanics of quantum systems. The quantities of interest are the

partition function of the system,

Z() = Tr eH (1)

5We should note that the contemporaneous explosion of work on the one-dimensional electron gas

(Emery, 1979) provided important, early illustrations of these ideas.

6


7/62

and the expectation values of various operators,

O = 1Z()

Tr OeH. (2)

In writing these formal expressions we have assumed a finite temperature, kBT = 1/. To get

at what happens exactly at T = 0 we take the T 0 limit. Upon doing so, the free energy,F = 1 ln Z(), becomes the ground state energy and the various thermal averages becomeground state expectation values. From Z we can get all the thermodynamic quantities of

interest. Expectation values of operators of the form O A(rt)A(rt) are related to theresults of dynamical scattering and linear response measurements. For example, A might

be the local density (X-ray scattering) or current (electrical transport).

A. Partition Functions and Path Integrals

Let us focus for now on the expression for Z. Notice that the operator density matrix,

eH, is the same as the time evolution operator, eiHT/h, provided we assign the imaginary

value T = ih to the time interval over which the system evolves. More precisely, whenthe trace is written in terms of a complete set of states,

Z() =n

n|eH|n, (3)

Z takes the form of a sum of imaginary time transition amplitudes for the system to start

in some state |n and return to the same state after an imaginary time interval ih. Thuswe see that calculating the thermodynamics of a quantum system is the same as calculating

transition amplitudes for its evolution in imaginary time, the total time interval being fixed

by the temperature of interest. The fact that the time interval happens to be imaginary is

not central. The key idea we hope to transmit to the reader is that Eq.(3) should evoke an

image of quantum dynamics and temporal propagation.

This way of looking at things can be given a particularly beautiful and practical imple-

mentation in the language of Feynmans path integral formulation of quantum mechanics

(Feynman, 1972). Feynmans prescription is that the net transition amplitude between two

7


8/62

states of the system can be calculated by summing amplitudes for all possible paths between

them. The path taken by the system is defined by specifying the state of the system at a

sequence of finely spaced intermediate time steps. Formally we write

eH =e

1hH

N

, (4)

where is a time interval6 which is small on the time scales of interest ( = h/ where

is some ultraviolet cutoff) and N is a large integer chosen so that N = h. We then

insert a sequence of sums over complete sets of intermediate states into the expression for

Z():

Z() =n

m1,m2,...,mN

n|e1

h H|m1m1|e1

h H|m2m2| . . . |mNmN|e1

h H|n. (5)

This rather messy expression actually has a rather simple physical interpretation. Formally

inclined readers will observe that the expression for the quantumpartition function in Eq. (5)

has the form of a classical partition function, i.e. a sum over configurations expressed in

terms of a transfer matrix, if we think of imaginary time as an additional spatial dimension.

In particular, if our quantum system lives in d dimensions, the expression for its partition

function looks like a classical partition function for a system with d + 1 dimensions, except

that the extra dimension is finite in extenth in units of time. As T 0 the systemsize in this extra time direction diverges and we get a truly d + 1 dimensional, effective,

classical system.

Since this equivalence between a d dimensional quantum system and a d + 1 dimensional

classical system is crucial to everything else we have to say, and since Eq. (5) is probably not

very illuminating for readers not used to a daily regimen of transfer matrices, it will be very

useful to consider a specific example in order to be able to visualize what Eq. (5) means.

6For convenience we have chosen to be real, so that the small interval of imaginary time that

it represents is i.

8


9/62

B. Example: 1D Josephson Junction Arrays

Consider a one-dimensional array comprising a large number L of identical Josephson

junctions as illustrated in Fig. (1). Such arrays have recently been studied by by Haviland

and Delsing. (Haviland and Delsing, 1996) A Josephson junction is a tunnel junction con-

necting two superconducting metallic grains. Cooper pairs of electrons are able to tunnel

back and forth between the grains and hence communicate information about the quantum

state on each grain. If the Cooper pairs are able to move freely from grain to grain through-

out the array, the system is a superconductor. If the grains are very small however, it costs

a large charging energy to move an excess Cooper pair onto a grain. If this energy is large

enough, the Cooper pairs fail to propagate and become stuck on individual grains, causing

the system to be an insulator.

The essential degrees of freedom in this system are the phases of the complex super-

conducting order parameter on the metallic elements connected by the junctions7 and

their conjugate variables, the charges (excess Cooper pairs, or equivalently the voltages)

on each grain. The intermediate state |mj at time j j, that enters the quan-tum partition function Eq. (5), can thus be defined by specifying the set of phase angles

{(j)} [1(j), 2(j), . . . , L(j)]. Two typical paths or time histories on the interval[0, h] are illustrated in Fig. (2) and Fig. (3), where the orientation of the arrows (spins)

indicates the local phase angle of the order parameter. The statistical weight of a given

path, in the sum in Eq. (5), is given by the product of the matrix elements

j

{(j+1)}|e 1h H|{(j)}, (6)

where

H =C

2

j

V2j EJ cos

j j+1

, (7)

7It is believed that neglecting fluctuations in the magnitude of the order parameter is a good

approximation; see Bradley and Doniach (1984); Wallin et al. (1994)

9


10/62

is the quantum Hamiltonian of the Josephson junction array. Here j is the operator repre-

senting the phase of the superconducting order parameter on the jth grain8; Vj i2eC jis conjugate to the phase9 and is the voltage on the jth junction, and EJ is the Josephson

coupling energy. The two terms in the Hamiltonian then represent the charging energy of

each grain and the Josephson coupling of the phase across the junction between grains.

As indicated previously, we can map the quantum statistical mechanics of the array onto

classical statistical mechanics by identifying the the statistical weight of a space-time path in

Eq. (6) with the Boltzmann weight of a two-dimensional spatial configuration of a classical

system. In this case the classical system is therefore a two-dimensional X-Y model, i.e. its

degrees of freedom are planar spins, specified by angles i, that live on a two-dimensional

square lattice. (Recall that at temperatures above zero, the lattice has a finite width h/

in the temporal direction.) While the degrees of freedom are easily identified, finding the

classical hamiltonian for this X-Y model is somewhat more work and requires an explicit

evaluation of the matrix elements which interested readers can find in the Appendix.

It is shown in the Appendix that, in an approximation that preserves the universality

class of the problem10, the product of matrix elements in Eq. (6) can be rewritten in the

8Our notation here is that {()} refers to the configuration of the entire set of angle vari-

ables at time slice . The s appearing in the Hamiltonian in Eq.(7) are angular coordinate

operators and j is a site label. The state at time slice is an eigenfunction of these operators:

cos

j j+1

|{()} = cos (j() j+1()) |{()}.9It is useful to think of this as a quantum rotor model. The state with wave function eimjj has

mj units of angular momentum representing mj excess Cooper pairs on grain j. The Cooper-pair

number operator in this representation is nj = i j . See (Wallin, et al., 1994). The cosine term in

Eq.(7) is a torque term which transfers units of conserved angular momentum (Cooper pairs) from

site to site. Note that the potential energy of the bosons is represented, somewhat paradoxically,

by the kinetic energy of the quantum rotors and vice versa.10That is, the approximation is such that the universal aspects of the critical behavior such as

10


11/62

form eHXY where the Hamiltonian of the equivalent classical X-Y model is

HXY =1

K

ij

cos(i j), (8)

and the sum runs over near-neighbor points in the two-dimensional (space-time) lattice.11

The nearest neighbor character of the couplings identifies the classical model as the 2D

X-Y model, extensively studied in the context of superfluid and superconducting films

(Goldenfeld, 1992; Chaikin and Lubensky, 1995). We emphasize that while the straight-

forward identification of the degrees of freedom of the classical model in this example is

robust, this simplicity of the resulting classical Hamiltonian is something of a minor mira-

cle.

It is essential to note that the dimensionless coupling constant K in HXY, which plays

the role of the temperature in the classical model, depends on the ratio of the capacitive

charging energy EC =(2e)2

C to the Josephson coupling EJ in the array,

K

ECEJ

. (9)

and has nothing to do with the physical temperature. (See Appendix.) The physics here is

that a large Josephson coupling produces a small value of K which favors coherent ordering

of the phases. That is, small K makes it unlikely that i and j will differ significantly, even

when sites i and j are far apart (in space and/or time). Conversely, a large charging energy

leads to a large value of K which favors zero-point fluctuations of the phases and disorders

the system. That is, large K means that the s are nearly independent and all values are

nearly equally likely.12 Finally, we note that this equivalence generalizes to d-dimensional

the exponents and scaling functions will be given without error. However, non-universal quantitiessuch as the critical coupling will differ from an exact evaluation. Technically, the neglected terms

are irrelevant at the fixed point underlying the transition.11Notice this crucial change in notation from Eq.(7) where j refers to a point in 1D space, not

1+1D space-time.12Because particle number is conjugate to the phase [nj = i j ], a state of indefinite phase on

a site has definite charge on that site, as would be expected for an insulator.

11


12/62

arrays and d+1-dimensional classical XY models.

C. Quantum-Classical Analogies

This specific example of the equivalence between a quantum system and a classical system

with an extra temporal dimension, illustrates several general correspondences between

quantum systems and their effective classical analogs.

Standard lore tells us that the classical XY model has an order-disorder phase transition

as its temperature K is varied. It follows that the quantum array has a phase transition as

the ratio of its charging and Josephson energies is varied. One can thus see why it is said

that the superconductor-insulator quantum phase transition in a 1-dimensional Josephson

junction array is in the same universality class as the order-disorder phase transition of the

1+1-dimensional classical XY model. [One crucial caveat is that the XY model universality

class has strict particle-hole symmetry for the bosons (Cooper pairs) on each site. In reality,

Josephson junction arrays contain random offset charges which destroy this symmetry and

change the universality class (Wallin, et al., 1994), a fact which is all too often overlooked.]

We emphasize again that K is the temperature only in the effective classical problem. In

the quantum case, the physical temperature is presumed to be nearly zero and only enters as

the finite size of the system in the imaginary time direction. The coupling constant K, the

fake temperature, is a measure not of thermal fluctuations, but of the strength of quantum

fluctuations, or zero point motion of the phase variables.13 This notion is quite confusing, so

the reader might be well advised to pause here and contemplate it further. It may be useful

to examine Fig. (4), where we show a space time lattice for the XY model corresponding

to a Josephson junction array at a certain temperature, and at a temperature half as large.

13Zero point motion of the phase variables is caused by the fact that there is an uncertainty

relation between the phase and the number of Cooper pairs on a superconducting grain. The more

well-defined the phase is, the larger the uncertainty in the charge is. This costs capacitive energy.

12


13/62

The size of the lattice constant in the time direction [ in the path integral in Eq. (5)] and

K are the same in both cases even though the physical temperature is not the same. The

only difference is that one lattice is larger in the time direction than the other.

In developing intuition about this picture, it may be helpful to see how classical physics

is recovered at very high temperatures. In that limit, the time interval h is very short

compared to the periods associated with the natural frequency scales in the system and

typical time histories will consist of a single static configuration which is the same at each

time slice. The dynamics therefore drops out of the problem and a Boltzmann weight

exp(Hclassical) is recovered from the path integral.The thermodynamic phases of the array can be identified from those of the XY model.

A small value of K corresponds to low temperature in the classical system and so the

quantum system will be in the ordered ferromagnetic phase of the XY model, as illustrated

in Fig. (2). There will be long-range correlations in both space and time of the phase

variables.14 This indicates that the Josephson coupling dominates over the charging energy,

and the order parameter is not fluctuating wildly in space or time so that the system is in

the superconducting phase. For large K, the system is disordered and the order parameter

fluctuates wildly. The correlations decay exponentially in space and time as illustrated in

Fig. (3). This indicates that the system is in the insulatingphase, where the charging energy

dominates over the Josephson coupling energy.

Why can we assert that correlations which decay exponentially in imaginary time indicate

an excitation gap characteristic of an insulator? This is readily seen by noting that the

Heisenberg representation of an operator in imaginary time is

A() = eH /hAeH /h (10)

and so the (ground state) correlation function for any operator can be expressed in terms of

14In this special 1+1D case, the correlations happen to decay algebraically rather than being truly

of infinite range.

13


14/62

a complete set of states as

G() 0|A()A(0)|0 = m

e(m0)/h|0|A|m|2, (11)

where m is the energy of the mth excited state. The existence of a finite minimum excitation

gap 01 1 0 guarantees that for long (imaginary) times the correlation function willdecay exponentially,15 i.e.,

G() e01/h . (12)

To recapitulate, we have managed to map the finite temperature 1D quantum problem

into a 2D classical problem with one finite dimension that diverges as T

0. The parameter

that controls the fluctuations in the effective classical problem does not involve T, but

instead is a measure of the quantum fluctuations. The classical model exhibits two phases,

one ordered and one disordered. These correspond to the superconducting and insulating

phases in the quantum problem. In the former the zero-point or quantum fluctuations of

the order parameter are small. In the latter they are large. The set of analogies developed

here between quantum and classical critical systems is summarized in Table I.

Besides the beautiful formal properties of the analogy between the quantum path integral

and d + 1 dimensional statistical mechanics, there are very practical advantages to this

analogy. In many cases, particularly for systems without disorder, the universality class of

the quantum transition is one that has already been studied extensively classically and a

great deal may already be known about it. For new universality classes, it is possible to do

the quantum mechanics by classical Monte Carlo or molecular dynamics simulations of the

appropriate d + 1-dimensional model.

Finally, there is a special feature of our particular example that should be noted. In this

case the quantum system, the 1D Josephson junction array (which is also the 1D quantum

15At T = 0 and for very long times (comparable to h), the finiteness in the time direction will

modify this result. Also, we implicitly assume here that 0|A|0 = 0.

14


15/62

X-Y model), has mapped onto a classical model in which space and time enter in the same

fashion, i.e., the isotropic 2D classical X-Y model. Consequently, the dynamical exponent

z (to be defined below) is unity. This is not true in generaldepending upon the quantum

kinetics, the coupling in the time direction can have a very different form and the effective

classical system is then intrinsically anisotropic and not so simply related to the starting

quantum system.

D. Dynamics and Thermodynamics

We end this account of quantum statistical mechanics by commenting on the relation-

ship between dynamics and thermodynamics. In classical statistical mechanics, dynamics

and thermodynamics are separable, i.e., the momentum and position sums in the partition

function are totally independent. For example, we do not need to know the mass of the

particles to compute their positional correlations. In writing down simple non-dynamical

models, e.g. the Ising model, we typically take advantage of this simplicity.

This freedom is lost in the quantum problem because coordinates and momenta do not

commute.16 It is for this reason that our path integral expression for Z contains information

on the imaginary time evolution of the system over the interval [0, h], and, with a little

bit of care, that information can be used to get the dynamics in real time by the analytic

continuation,

16Stated more formally, calculating Z classically only requires knowledge of the form of the Hamil-

tonian function and not of the equations of motion, while both enter the quantum calculation.

Recall that H alone does not fix the equations of motion; one also needs the Poisson brack-

ets/commutators among the phase space variables. While these determine the classical oscillation

frequencies, they do not enter the classical calculation of Z. In quantum mechanics h times the

classical oscillation frequencies yields the energy level spacing. Hence the commutators are needed

to find the eigenvalues of the quantum H needed to compute the quantum Z.

15


16/62

G() G(+it) (13)

in Eq. (11). Stating it in reverse, one cannot solve for the thermodynamics without also solv-

ing for the dynamicsa feature that makes quantum statistical mechanics more interesting

but that much harder to do!

Heuristically, the existence of h implies that energy scales that enter thermodynamics

necessarily determine time scales which then enter the dynamics and vice-versa. Consider the

effect of a characteristic energy scale, such as a gap , in the spectrum. By the uncertainty

principle there will be virtual excitations across this gap on a time scale h/, which will

appear as the characteristic time scale for the dynamics. Close to the critical point, where

vanishes, and at finite temperature this argument gets modifiedthe relevant uncertainty

in the energy is now kBT and the characteristic time scale is h. In either case, the linkage

between dynamics and thermodynamics is clear.

II. QUANTUM PHASE TRANSITIONS

We now turn our attention to the immediate neighborhood of a quantum critical point.

In this region the mapping of the quantum system to a d+1 dimensional classical model will

allow us to make powerful general statements about the former using the extensive lore on

critical behavior in the latter. Hence most of the following will consist of a reinterpretation

of standard ideas in classical statistical mechanics in terms appropriate for d +1 dimensions,

where the extra dimension is imaginary time.

A. T = 0: Dynamic Scaling

In the vicinity of a continuous quantum phase transition we will find several features of

interest. First, we will find a correlation length that diverges as the transition is approached.

That diverging correlation lengths are a generic feature of classical critical points, immedi-

ately tells us that diverging lengths and diverging times are automatically a generic feature

16


17/62

of quantum critical points, since one of the directions in the d+1 dimensional space is time.

This makes sense from the point of view of causality. It should take a longer and longer time

to propagate information across the distance of the correlation length.

Actually, we have to be carefulas we remarked earlier, the time direction might easily

involve a different set of interactions than the spatial directions, leading to a distinct cor-

relation length in the time direction. We will call the latter , reserving the symbol

for the spatial correlation length. Generically, at T = 0 both (K) and (K) diverge as

K Kc 0 in the manner,17

||

z. (14)

These asymptotic forms serve to define the correlation length exponent , and the dynamical

scaling exponent, z. The nomenclature is historical, referring to the extension of scaling

ideas from the study of static classical critical phenomena to dynamics in the critical region

associated with critical slowing down (Hohenberg and Halperin, 1977; Ferrell, 1968). In

the classical problem the extension was a non-trivial step, deserving of a proper label. As

remarked before, the quantum problem involves statics and dynamics on the same footing

and so nothing less is possible. For the case of the Josephson junction array considered

previously, we found the simplest possible result, z = 1. As noted however this is a special

isotropic case and in general, z = 1.As a consequence of the diverging and , it turns out that various physical quantities in

the critical region close to the transition have (dynamic) scaling forms, i.e. their dependence

on the independent variables involves homogeneity relations of the form:

O(k,,K) = dOO(k,) (15)

17Here and in the following, we do not write the microscopic length and time scales that are

needed to make dimensional sense of these equations. See Goldenfeld (1992).

17


18/62

where dO is called the scaling dimension18 of the observable O measured at wavevector k

and frequency . The meaning of (and assumption behind) these scaling forms is simply

that, close to the critical point, there is no characteristic length scale other than itself19

and no characteristic time scale other than . Thus the specific value of the coupling K

does not appear explicitly on the RHS of Eq.(15). It is present only implicitly through the

K dependence of and .

If we specialize to the scale invariant critical point, the scaling form in Eq.(15) is no

longer applicable since the correlation length and times have diverged to infinity. In this

case the only characteristic length left is the wavelength 2/k at which the measurement

is being made, whence the only characteristic frequency is

kz. As a result we find the

simpler scaling form:

O(k,,Kc) = kdOO(kz/), (16)

reflecting the presence of quantum fluctuations on all length and time scales.20

The utility and power of these scaling forms can be illustrated by the following example.

In an ordinary classical system at a critical point in d dimensions where the correlation

length has diverged, the correlations of many operators typically fall off as a power law

G(r) O(r)O(0) 1r(d2+d)

, (17)

18The scaling dimension describes how physical quantities change under a renormalization group

transformation in which short wavelength degrees of freedom are integrated out. As this is partly

a naive change of scale, the scaling dimension is often close to the naive (engineering) dimen-

sion of the observable but (except at special, non-generic, fixed points) most operators develop

anomalous dimensions. See Goldenfeld (1992).19For a more precise statement that includes the role of cutoff scales, see Goldenfeld (1992).20Equivalently, we could have argued that the scaling function on the RHS of Eq.(15) must for

large arguments x, y have the form O(x, y) xdOO(xzy1) in order for the observable to have a

sensible limit as the critical point is approached.

18


19/62

so that the Fourier transform diverges at small wavevectors like

G(k) k2+d . (18)

Suppose that we are interested in a QPT for which the d+1-dimensional classical system is

effectively isotropic and the dynamical exponent z = 1. Then the Fourier transform of the

correlation function for the d + 1-dimensional problem is

G(k, n)

k2 + 2n

2+d+1, (19)

where the d + 1 component of the wavevector is simply the Matsubara frequency used to

Fourier transform in the time direction. Analytic continuation to real frequencies via the

usual prescription (Mahan, 1990) in + i yields the retarded correlation functionGR(k, + i)

k2 ( + i)2

(2+d+1)/2. (20)

Instead of a pole at the frequency of some coherently oscillating collective mode, we see

instead that GR(k, + i) has a branch cut for frequencies above = k (we have implicitly

set the characteristic velocity to unity). Thus we see that there is no characteristic frequency

other than k itself (in general, kz as in Eq.(16)), as we discussed above. This implies that

collective modes have become overdamped and the system is in an incoherent diffusive

regime. The review by Sachdev contains some specific examples which nicely illustrate

these points (Sachdev, 1996).

Finally, three comments are in order. First, as we saw in the example of the Josephson

junction array, a finite temporal correlation length means that there is a gap in the spectrum

of the quantum problem. Conversely, critical systems are gapless. Second, the exponent z

is a measure of how skewed time is, relative to space, in thecritical

region. This does not,a priori, have anything to do with what happens in either of the phases. For example, one

should resist the temptation to deduce the value of z via qz from the dispersion of anyGoldstone mode21 in the ordered phase. This is incorrect since the exponent z is a property

21A Goldstone mode is a gapless excitation that is present as a result of a broken continuous

19


20/62

of the critical point itself, not of the ordered phase. Third, we should restate the well known

wisdom that the diverging lengths and the associated scaling of physical quantities are

particularly interesting because they represent universal behavior, i.e., behavior insensitive

to microscopic details within certain global constraints such as symmetry and dimensionality

(Goldenfeld, 1992).

B. T = 0: Finite Size Scaling

So far we have described the framework, appropriate to the system at T = 0, that would

describe the underlying QPT in any system. As the experimentally accessible behavior of the

system necessarily involves a non-zero temperature, we need to understand how to modify

the scaling forms of the previous section for the T = 0 problem.The crucial observation for this purpose is , as noted earlier and illustrated in Fig. (4),

that the only effect of taking T = 0 in the partition function (5) is to make the temporaldimension finite; in particular, there is no change in the coupling K with physical temper-

ature. The effective classical system now resembles a hyper-slab with d spatial dimensions

(taken to be infinite in extent) and one temporal dimension of size L

h. As phase tran-

sitions depend sensitively upon the dimensionality of the system, we expect the finiteness

of L to modify the critical behavior, since at the longest length scales the system is now d

dimensional.

This modification can take two forms. First, it can destroy the transition entirely so that

symmetry in the ordered phase of a system. Broken continuous symmetry means that the energy

is degenerate under a continuous family of uniform global symmetry transformations, for example

uniform rotation of the magnetization in an XY magnet. This implies that non-uniform but long

wavelength rotations must cost very little energy and hence there exists a low-energy collective

mode in which the order parameter fluctuates at long wavelengths. See Goldenfeld (1992) and

Chaikin and Lubensky (1995).

20


21/62

the only critical point is at T = 0. This happens in the case of the 1D Josephson array. Its

finite temperature physics is that of an XY model on an infinite strip which, being a one

dimensional classical system with short-range forces, is disordered at all finite values of K

(finite temperatures in the classical language).

In the second form, the modification is such that the transition persists to T = 0 butcrosses over to a different universality class. For example, the problem of a 2D Josephson

junction array maps onto a 3(=2+1) dimensional classical XY model. Its phase diagram is

illustrated in Fig. (5). At T = 0 the QPT for the transition from superconductor to insulator

is characterized by the exponents of the 3D XY model. That is, it looks just like the classical

lambda transition in liquid helium with K

Kc playing the role of T

Tc in the helium.

However at T = 0 the system is effectively two dimensional and undergoes a transition of the2D Kosterlitz-Thouless22 XY variety at a new, smaller, value of K, much like a helium film.

The Kosterlitz-Thouless (KT) transition occurs on the solid line in Fig. (5). We see that it

is necessary to reduce the quantum fluctuations (by making K smaller) in order to allow the

system to order at finite temperatures. Above some critical temperature, the system will

become disordered (via the KT mechanism) owing to thermal fluctuations. Of course, if we

make K larger the quantum fluctuations are then so large that the system does not order

at any temperature, even zero. The region on the K axis (i.e., at T = 0) to the right of the

QCP in Fig. (5) represents the quantum disordered superconductor, that is, the insulator.

At finite temperatures, no system with a finite gap is ever truly insulating. However there

is a crossover regime, illustrated by the dotted line in Fig. (5) separating the regimes where

the temperature is smaller and larger than the gap.

At this point the reader might wonder how one learns anything at all about the QPT if

the effects of temperature are so dramatic. The answer is that even though the finiteness of

22The Kosterlitz-Thouless phase transition is a special transition exhibited by two-dimensional

systems having a continuous XY symmetry. It involves the unbinding of topological vortex defects.

See Goldenfeld (1992) and Chaikin and Lubensky (1995).

21


22/62

L causes a crossover away from the T = 0 behavior, sufficiently close to the T = 0 critical

point, it does so in a fashion controlled by the physics at that critical point. This is not

an unfamiliar assertion. In the language of the renormalization group, critical points are

always unstable fixed points and lead to scaling not because they decide where the system

ends up but because they control how long it takes to get there. Here, instead of moving

the system away from the critical fixed point by tuning a parameter, we do so by reducing

its dimensionality.

Since the physics has to be continuous in temperature, the question arises of how large

the temperature has to be before the system knows that its dimension has been reduced.

The answer to this is illustrated in Fig. (6). When the coupling K is far away from the zero-

temperature critical coupling Kc the correlation length is not large and the corresponding

correlation time z is small. As long as the correlation time is smaller than thesystem thickness h, the system does not realize that the temperature is finite. That is,

the characteristic fluctuation frequencies obey h kBT and so are quantum mechanicalin nature. However as the critical coupling is approached, the correlation time grows and

eventually exceeds h. (More precisely, the correlation time that the system would have had

at zero temperature exceeds h; the actual fluctuation correlation time is thus limited by

temperature.) At this point the system knows that the temperature is finite and realizes

that it is now effectively a d-dimensional classical system rather than a d + 1-dimensional

system.

The formal theory of the effect of reduced dimensionality near critical points, which

quantifies the above considerations, is called finite size scaling (Privman, 1990). For our

problem it asserts that for [K

Kc]/K

c 1 and T

0, physical quantities have the finite

size scaling form,

O(k,,K,T ) = LdO/z O(kL1/z , L, L/). (21)

The interpretation of this form is the following. The quantity L h defined above leads,as discussed in more detail below, to a characteristic length L1/z associated with the

22


23/62

temperature. Hence the prefactor LdO/z is the analog of the corresponding prefactor in

Eq. (15). This same characteristic length is the only one against which to measure the wave

vector k. The associated time L is the time scale against which to measure the frequency in

the second term. Finally the distance to the zero temperature critical coupling is measured

via the ratio of L to the zero temperature correlation time . The message here is that

what matters is the ratio of the finite size in the time direction to the T = 0 correlation

length in that direction. We will return to the uses of this form shortly.

Our considerations in this section also show us why the phase boundary in Fig. (5)

(solid line) and the crossover line (dashed line) reach zero temperature in a singular way

as the quantum critical point is approached. Simple dimensional analysis tells us that the

characteristic energy scale ( for the insulator, TKT for the superfluid) vanishes near the

critical point like h/, implying

|K Kc|z(K Kc)

TKT |K Kc|z(Kc K). (22)

C. The Quantum-Classical Crossover and the Dephasing Length

We now turn to a somewhat different understanding and interpretation of the effect of

temperature that is conceptually of great importance. Recall that the T = 0 critical points

of interest to us are gapless and scale invariant, i.e., they have quantum fluctuations at

all frequencies down to zero. Temperature introduces a new energy scale into the prob-

lem. Modes whose frequencies are larger than kBT /h are largely unaffected, while those

with frequencies lower than kBT /h become occupied by many quanta with the consequence

that they behave classically. Put differently, the temperature cuts off coherent quantum

fluctuations in the infrared.

What we want to show next is that this existence of a quantum to classical crossover

frequency (kBT /h) leads to an associated length scale for the same crossover, as alluded

23


24/62

to in the previous section. We shall refer to this length scale as the dephasing length, L,

associated with the QPT. The temperature dependence of L is easy enough to calculate.

From our imaginary time formalism we recall that quantum fluctuations are fluctuations

in the temporal direction. Evidently these cannot be longer-ranged than the size of the

system in the time direction, L = h. Since spatial and temporal correlations are linked

via z, it follows that the spatial correlations linked with quantum fluctuations are notlonger ranged than L1/z . Since the spatial range of quantum fluctuations is the dephasing

length, we find L 1/T1/z.We use the term dephasing deliberately. Readers may know, from other contexts

where quantum effects are observed, that such observation requires phase coherence for the

relevant degrees of freedom. In other words, interference terms should not be wiped out

by interactions with an environment, i.e. other degrees of freedom (Stern, et al., 1990).

If dephasing takes place and the phase coherence is lost, the resulting behavior is classical.

Thus our definition of L is in line with that notion. However, readers familiar with the

notion of a dephasing length, , in mesoscopic condensed matter physics or the theory of

Anderson localization, might be concerned at our appropriation of this notion. The concern

arises because, in the standard lore in those fields, one is often dealing with models of non-

interacting or even interacting electrons, whose quantum coherence is limited by degrees of

freedom, e.g. phonons, that are not being considered explicitly. This has given rise to a

tradition of thinking of as being a length which is set externally. Unfortunately this sort

of separation of degrees of freedom should not be expected to work near a QPT, since there

one needs to keep track of all relevant interactions. If a given interaction, e.g. the Coulomb

interaction, is relevant, then it already sits in the Hamiltonian we need to solve and enters

the calculation of L. In contrast, if an interaction, e.g. coupling to phonons, is irrelevant

then we do not expect it to enter the low energy physics as it should not in general affect

the quantum-classical crossover either.

Another way of formulating this is in terms of dephasing rates. Since temperature is

the only energy scale available, a generic quantum critical point will be characterized by a

24


25/62

dephasing rate 1 that is linear in T, since we expect h/ kBT. By definition, irrelevantinteractions, e.g. phonons, have an effective coupling to the electrons which scales away to

zero as one examines the system on larger length and time scales. Hence such couplings

will produce a dephasing rate which vanishes as Tp

with p > 1 and will therefore become

negligible compared to T at low temperatures.23

Thus we conclude that L is the unique dephasing length in the vicinity of a generic

quantum critical point. Further discussion and explicit examples of the dissipative dynamics

near a critical point can be found in the article by Sachdev (1996).

III. EXPERIMENTS: QPTS IN QUANTUM HALL SYSTEMS

We now turn from our somewhat abstract considerations to examples of how one actually

looks for a QPT in experiments. The basic idea is relatively simple. We try to arrange that

the system is close to the conjectured critical point in parameter space (K Kc) and tem-perature (T 0) and look for mutually consistent evidence of scaling with various relevantparameters. By these we mean either the deviation of the quantum coupling constant from

its critical value K

Kc, the temperature, or the wavevector, frequency and amplitude of a

probe. We call these relevant parameters since when they are non-zero the response of the

system has no critical singularities due to the quantum critical pointhence the analogy

23Equivalently, the associated length scale will diverge faster than L as T 0 and hence will not

control the quantum-classical crossover of the relevant degrees of freedom, since it is the shortest

length that counts. There are times when this sort of reasoning can break down. These situations

involve operators that are irrelevant, i.e., decrease in the infrared, but cannot be naively set to

zero since that produces extra singularities. Such operators are known in the trade as dangerous

irrelevant operators and we will meet an example in the next section, in the context of current

scaling.

25


26/62

to relevant operators in renormalization group theory.24 Additionally, we can look for uni-

versal critical amplitudes or amplitude ratios that are implicit in scaling forms for various

quantities.

To see how this works, we will consider as a specific example, a set of recent experi-

ments on phase transitions in quantum Hall systems. We derive various scaling relations

appropriate to the experiments, even though we do not actually know the correct theory

describing the QPT in this disordered, interacting fermi system. The very generality of the

scaling arguments we will apply implies that one need not have a detailed microscopic under-

standing of the physics to extract useful information. Nevertheless, we will start with some

introductory background for readers unfamiliar with the quantum Hall effect (Prange and

Girvin, 1990; Chakraborty and Pietilainen, 1995; MacDonald, 1989; Stone, 1992; Kivelson

et al., 1993; Das Sarma and Pinczuk, 1996).

The quantum Hall effect (QHE) is a property of a two dimensional electron system placed

in a strong transverse magnetic field, B 10T. These systems are produced, using modernsemiconductor fabrication techniques, at the interface of two semiconductors with different

band gaps. The electronic motion perpendicular to the interface is confined to a potential

well 100A thick. Because the well is so thin, the minimum excitation energy perpendicular

24Consider for example a weak magnetic field applied to a system undergoing ferromagnetic or-

dering. The magnetic field is relevant and removes the sharp singularity in magnetization at the

critical temperature, replacing it with a rapid but smooth increase in magnetization. Likewise,

measurements at a non-zero frequency and wavevector do not exhibit singularities across a transi-

tion. For quantum systems, changes in the coupling and the temperature can produce more subtle

effects: they will cut off the critical fluctuations coming from the proximity to the quantum critical

point, but either Goldstone modes coming from a broken continuous symmetry or purely classical

(thermal) fluctuations could lead to independent singularities in the thermodynamics and response.

We saw an example of the latter in the persistence of a phase transition at finite temperatures for

the 2D Josephson junction array.

26


27/62

to the 2D plane ( 200K) is much larger than the temperature ( 1K) and so motion inthis third dimension is frozen out, leaving the system dynamically two-dimensional.

As the ratio of the density to the magnetic field is varied at T = 0, the electrons

condense into a remarkable sequence of distinct thermodynamic phases.25

These phases are

most strikingly characterized by their unique electrical transport properties, as illustrated

in Fig (7). Within each phase the current flow is dissipationless, in that the longitudinal

resistivity, L, that gives the electric field along the direction of current flow (EL = Lj)

vanishes. At the same time that the longitudinal resistivity vanishes, the Hall resistivity, H,

that gives the electric field transverse to the direction of current flow (EH = Hj) becomes

quantized in rational multiples of the quantum of resistance

H =h

Be2(23)

where B is an integer or simple rational fraction which serves to distinguish between the

different phases. This quantization has been verified to an accuracy of about one part in

107 and to an even higher precision.

The QHE arises from a commensuration between electron density and magnetic flux

density, i.e. a sharp lowering of the energy when their ratio, the filling factor B, takes on

particular rational values. This commensuration is equivalent to the existence of an energy

gap in the excitation spectrum at these magic filling factors and leads to dissipationless

flow at T = 0 since degrading a small current requires making arbitrarily small energy excita-

tions which are unavailable. In the absence of disorder, Eqn. (23) follows straightforwardly,

for example by invoking Galilean invariance (Prange and Girvin, 1990). As the magnetic

field (or density) is varied away from the magic filling factor, it is no longer possible for the

system to maintain commensuration over its entire area, and it is forced to introduce a cer-

tain density of defects to take up the discommensuration. In the presence of disorder, these

25The exact membership of the sequence is sample specific but obeys certain selection rules

(Kivelson et al., 1992).

27


28/62

defects, which are the quasiparticles of the system, become localized and do not contribute

to the resistivities, which remain at their magic values. In this fashion, we get a QH phase

or a plateau.

Transitions between QH phases occur when too many quasiparticles have been intro-

duced into the original QH state and it becomes energetically favorable to move to the

basin of attraction of a different state and its associated defects. It might appear that

these transitions between neighboring phases are first order, since H jumps discontinuously

by a discrete amount between them, but they are not. Qualitatively, they involve the quasi-

particles of each phase which are localized on a length scale, the localization length, that

diverges as the transition is approached from either side. However, as these quasiparticles

are always localized at the longest length scale away from criticality, they do not lead to

dissipation (L = 0) and do not renormalize the Hall resistivities of their respective phases.

Exactly at the transition they are delocalized and lead to a non-zero L. The shift in H

on moving through the transition can be understood in terms of either set of quasiparticles

condensing into a fluid statethere being an underlying duality in this description.

In our description of the QH phases and phase transitions we have employed a common

language for all of them. We should note that this does not, ipso facto imply that all quantum

Hall transitions are in the same universality class; however, experiments, as we discuss later,

do seem to suggest that conclusion. The reason for this caution is that different QH states

can arise from quite different physics at the microscopic level. States with integer B arise, to

first approximation, from single particle physics. An electron in a plane can occupy a Landau

level which comprises a set of degenerate states with energy (n + 1/2)hc; these reflect the

quantization of the classical cyclotron motion having frequency c

= eBm

and the arbitrariness

of the location of that motion in the plane. When an integer number of Landau levels are

full, and this corresponds to an integer filling factor, excitations involve the promotion of an

electron across the cyclotron gap and we have the commensuration/gap nexus necessary for

the observation of the (integer) QHE. In contrast, fractional filling factors imply fractional

occupations of the Landau levels, with attendant macrosopic degeneracies, and they exhibit

28


29/62

a gap only when the Coulomb interaction between the electrons is taken into account (the

fractional QHE).26 Readers interested in the details of this magic trick are encouraged to

peruse the literature.

Before proceeding to the details of experiments, we need to discuss two important points

about the units of the quantities measured in electrical transport where two spatial dimen-

sions are rather special. Experiments measure resistances, which are ratios of total voltages

to current and these are related to local resistivities by ratios of cross-sectional areas to

lengths. In two dimensions, a cross-sectional area is a length and consequently no factor

of length intervenes between the global and local quantities. In the QH phases, this has

the important implication that no geometrical factor affects the measurement of the Hall

resistance, which is why the ratio of fundamental constants h/e2 and hence the fine struc-

ture constant can be measured to high accuracy on samples whose geometry is certainly not

known to an accuracy of one part in 107.

What we have said above is a statement about the engineering dimensions of resistance

and resistivity. Remarkably, this also has an analog when it comes to their scaling dimen-

sions at a quantum critical point, i.e. their scaling dimensions vanish.27 Consequently, the

resistivities vary as the zeroth power of the diverging correlation length on approaching

the transition, i.e. will remain constant on either side. Precisely at criticality they will be

independent of the length scale used to define them but can take values distinct from the

neighboring phases.

In this fashion, we have recovered from a purely scaling argument our earlier conclusion

26This leads to the remarkable feature that while the quasiparticles of the integer states are

essentially electrons, those of the fractional states are fractionally charged and obey fractional

statistics.27See (Fisher, et al., 1990; Cha, et al., 1991). This is analogous to the behavior of the superfluid

density at the classical Kosterlitz-Thouless phase transition (Chaikin and Lubensky, 1995) and

leads to a universal jump in it.

29


30/62

that even though the quantum Hall transitions are continuous, the resistivities at T = 0

differ from the quantized values only at critical points. Detecting the continuous transitions

then requires measurements at a non-zero temperature, frequency or current, all of which

lead to a more gradual variation which can then be examined for scaling behavior. Below

are some examples of how that works.

A. Temperature and Frequency Scaling

Consider the caricature of a typical set of data shown in Fig. (8). Note that H inter-

polates between its quantized values over a transition region of non-zero width, while L is

peaked in the same region, but is extremely small outside the region. The change in the

shape of these curves with temperature can be understood on the basis of the finite size

scaling form

L/H(B , T , ) = fL/H(h/kBT,/T1/z), (24)

where (B Bc)/Bc measures the distance to the zero temperature critical point. Thisform is equivalent to the general finite-size scaling form in Eq. (21) except that we have

assumed the limit k = 0, and used the previously cited result that the scaling dimension of

the resistivity vanishes in d = 2 (Fisher, et al., 1990; Cha, et al., 1991). The first argument

in the scaling function here is the same as the second in Eq. (21). The second argument in

the scaling function here is simply a power of the third argument in Eq. (21). This change

is inconsequential; it can be simply absorbed into a redefinition of the scaling function.

First, let us consider a DC or = 0 measurement. In this case our scaling form implies

that the resistivities are not independent functions of (or B) and T but instead are functions

of the single scaling variable /T1/z . Hence the effect of lowering T is to rescale the deviation

of the field from its critical value by the factor T1/z . It follows that the transition appears

sharper and sharper as the temperature is lowered, its width vanishing as a universal power

of the temperature, B T1/z . In Fig. (9) we show the pioneering data of Wei et al.(1988) that indeed shows such an algebraic dependence for several different transitions all

30


31/62

of which yield the value 1/z 0.42. These transitions are between integer quantum Hallstates. Remarkably, this temperature scaling behavior seems to be ubiquitous at quantum

Hall transitions and suggests that there is a single underlying fixed point for all of them.

It was shown by Engel et al. (1990) that it holds at transitions between two fractional

quantum Hall states. Subsequently, Wong et al. (1995) found the same scaling for the

transition between a Hall state and the insulator. In Fig. (10) we show some recent data of

Shahar (1995) near another such transition, plotted both as a function of the magnetic field

at several values ofT, and against the scaling variable /T1/z , exhibiting the data collapse

characteristic of the critical region.

Consider now the results of measurements at non-zero frequencies. In their full generality

these require a two variable scaling analysis (Engel, et al., 1995) but we focus instead on

two distinct regimes. In the regime h kBT we expect that the behavior of the scalingfunction will be governed by its = 0 limit analyzed previously, i.e. at small we expect

the scaling to be dominated by T. In the second regime, h kBT, we expect the scalingto be dominated by and the scaling function to be independent of T. In order for the

temperature to drop out, the scaling function in Eq. (24) must have the form

f(x, y) f(yx1/z) (25)

for large x so that the scaling variables conspire to appear in the combination,

h

kBT

1/z

T1/z

1/z. (26)

It follows that at high frequencies the resistivities are functions of the scaling variable /1/z

and that the width of the transition regions scales as 1/z . Fig. (11) shows frequency

dependent conductivity28 data of Engel et al. (1993) which exhibits this algebraic increase

in the width of the transition region with frequency and yields a value of z consistent with

the temperature scaling.

28The conductivities scale in exactly the same fashion as the resistivities.

31


32/62

We should note an important point here. As the ratio h/kBT is varied in a given

experiment we expect to see a crossover between the T and dominated scaling regimes.

The criterion for this crossover is h kBT. The observation by Engel et al. (1990), that

this is indeed the correct crossover criterion (see Fig. (11)) is important for two reasons.

First, it involves h and clearly implies that quantum effects are at issue. Second it implies

that T is the relevant infrared scale. If dephasing effects coming from coupling to some

irrelevant degree of freedom were important, one would expect the crossover to take place

when 1, where 1/ is some microscopic scattering or relaxation rate associated withthis coupling. Since the coupling is irrelevant it will, as noted earlier, give a scattering

rate that vanishes as ATp where p is greater than unity and A is non-universal (Sondhi

and Kivelson, 1992) (e.g., it depends on the precise value of the electron-phonon coupling

constant for the material). In contrast, what is observed is that the relaxation rate obeys

1/ = CkBT /h where C is a universal(Sachdev, 1996) dimensionless constant of order unity.

It is important to note that frequency scaling does not give us any new information on

exponents that we did not already have from the temperature scaling. The main import of

frequency scaling is its ability to confirm the quantum critical nature of the transition by

showing that the characteristic time scales have diverged, leaving the temperature itself as

the only frequency scale.

B. Current Scaling

A third relevant parameter that is experimentally useful is the magnitude of the mea-

suring current or, equivalently, of the applied electric field. In talking about resistivities we

have assumed that there is an ohmic regime at small currents, i.e., a regime in which the

voltages are linear in the current. In general, there is no reason to believe that the non-linear

response can be related to equilibrium propertiesi.e., there is no fluctuation-dissipation

theorem beyond the linear regime. However, in the vicinity of a critical point we expect the

dominant non-linearities to come from critical fluctuations. At T = 0, the electric field scale

32


33/62

for these can be estimated from an essentially dimensional argument. We start by defining

a characteristic length E associated with the electric field. Imagine that the system is at

the critical point so that E is the only length scale available. Then the only characteristic

time for fluctuations of the system will scale like +z

E . We can relate the length E to the

electric field that produces it by

eEE hzE . (27)

This expression simply equates the energy gained from the electric field by an electron

moving a distance E to the characteristic energy of the equilibrium system at that same

scale. Thus

E E1/(1+z). (28)

If the system is not precisely at the critical point, then it is this length E that we should

compare to the correlation length

E

E1/(1+z)

E1/(z+1)

. (29)

From this we find that the non-linear DC resistivities for a 2D system obey the scaling forms

L/H(B , T , E ) = gL/H(/T1/z

,/E1/(z+1)

). (30)

This is a very useful result because it tells us that electric field scaling will give us new

information not available from temperature scaling alone. From temperature scaling we can

measure the combination of exponents z . Because an electric field requires multiplication

by one power of the correlation length to convert it to a temperature (energy), electric field

scaling measures the combination of exponents (z +1). Thus the two measurements can be

combined to separately determine and z. The data of Wei et al. (1994), Fig. (12), confirm

this and yield the value (z + 1) 4.6. Together the T, and I scaling experiments leadto the assignment 2.3 and z 1.

Equation (30) tells us that there are two scaling regimes. At sufficiently high tem-

peratures, L E and the scaling is controlled by the temperature. Below a crossovertemperature scale

33


34/62

T0(E) zE Ez/(1+z), (31)

L > E and the scaling is controlled by the electric field E (or equivalently, the applied

current I). One might be tempted to identify T0(E) as the effective temperature of the

electrons in the presence of the electric field, but this is not strictly appropriate since the

system is assumed to have been driven out of equilibrium on length scales larger than E.

This quantum critical scaling picture explicitly assumes that the slow internal time scales

of the system near its critical point control the response to the electric field and implicitly

assumes that we can ignore the time scale which determines how fast the Joule heat can

be removed by phonon radiation. Thus this picture is quite distinct from that of a simple

heating scenario in which the electron gas itself equilibrates rapidly, but undergoes a rise in

temperature if there is a bottleneck for the energy deposited by Joule heating to be carried

away by the phonons. This effect can give rise to an apparent non-linear response that is, in

fact, the linear response of the electron gas at the higher temperature. The power radiated

into phonons at low electron temperatures scales as

Pph = ATe , (32)

where = 4 7 depending on details (Chow, et al., 1996). Equating this to the Jouleheating (assuming a scale invariant conductivity) yields an electronic temperature

Telec E2/. (33)

We now have a paradox. The more irrelevant phonons are at low temperatures (i.e., the

larger is), the smaller is the exponent 2/ and hence the more singular is the temperature

rise produced by the Joule heat. Comparing Eqs.(31) and (33) we see that for

2

T0(E). (35)

34


35/62

That is, we then have that the temperature rise needed to radiate away sufficient power

is larger than the characteristic energy (temperature) scale predicted by the quantum

critical scaling picture. In this case the phonons are dangerously irrelevant and the simple

quantum critical scaling prediction fails. It happens that for the case of GaAs, which is

piezoelectric, 2/ = 1/2 which gives the same singularity exponent as the quantum critical

model z/(z + 1) = 1/2 (since z = 1). Hence both quantum critical and heating effects are

important. (The phonon coupling is marginally dangerous.) This result is discussed in

more detail elsewhere (Chow, et al., 1996; Girvin, et al., 1996).

C. Universal Resistivities

The final signatures of critical behavior which we wish to discuss are universal amplitudes,

and, more generally, amplitude ratios. These are readily illustrated in the quantum Hall

problem without considering their general setting, for which we direct the reader to the

literature (Hohenberg, et al., 1976; Aharony and Hohenberg, 1976; Kim and Weichman,

1991; Chubukov and Sachdev, 1993; Sachdev, 1996). Note that the scaling forms (24) and

(30) imply that the resistivities at B = Bc in the critical region are independent of T,

and I. Under certain assumptions it is possible to argue that they are, in fact, universal

(Kivelson et al., 1992; Fisher, et al., 1990). The observation of such universality between

microscopically different samples would then be strong evidence for an underlying QPT as

well.

Recently Shahar et al. (1995) have carried out a study of the critical resistivities at the

transition from the B = 1 and 1/3 quantum Hall states to the insulating state. An example

of their data is shown in Fig. (10). Notice that there exists a critical value of B field at

which the resistivity is temperature-independent. For B < Bc the resistivity scales upward

with decreasing T, while for B > Bc, it scales downward with decreasing T. Since we can

think of lowering T as increasing the characteristic length scale L at which we examine the

system, we see that the point where all the curves cross is the scale-invariant point of the

35


36/62

system and hence must be the critical point.

Shahar et al. (1995) find that at these critical points, L is independent of the sample

studied and in fact appears to be h/e2 within experimental error for both transitions. Pre-

liminary studies (Shahar, 1995) also seem to find sample-independent values of H at the

critical points with values of h/e2 and 3h/e2 for the two transitions.

D. Unresolved Issues

As we have tried to indicate, the success of experimental work in making a case for

universal critical behavior at transitions in the quantum Hall regime is impressive. However,

not everything is settled on this score. Apart from the delicate issues surrounding the

interpretation of the current scaling data mentioned earlier, there is one significant puzzle.

This concerns the failure of L at the transition between two generic QH states to exhibit

a T-independent value at a critical field even as the width of the curve exhibits algebraic

behavior.29 This is generally believed to stem from macroscopic inhomogeneities in the

density and some recent theoretical work offers support for this notion (Ruzin, et al., 1996).

Nevertheless, this is an issue that will need further work. The transitions to the insulator

studied more recently, are believed to be much less sensitive to this problem, and hence the

consistency of the data on those is encouraging. However, in these cases the temperature

range over which there is evidence for quantum critical scaling is quite small as in the data

in Fig. (10) which leads us to a general caveat.

Evidence for power laws and scaling should properly consist of overlapping data that

cover several decades in the parameters. The various power law dependences that we have

exhibited span at best two decades, most of them fewer and the evidence for data collapse

within the error bars of the data exists only over a small range of the scaled variables.

Consequently, though the overall picture of the different types of data is highly suggestive,

29Hence our unwillingness to plot the actual traces in Fig. (8).

36


37/62

it cannot really be said that it does more than indicate consistency with the scaling expected

near a quantum critical point. Regrettably, there is at present no example of a quantum

critical phase transition as clean as the remarkable case of the classical lambda transition in

superfluid helium for which superb scaling can be demonstrated. (Ahlers, 1980)

On the theoretical front the news is mixed. Remarkably, the experimental value of the

correlation length exponent 2.3 is consistent with numerical calculations of the behaviorof non-interactingelectrons in a strong magnetic field (Huckestein, 1995). Also, the critical

resistivities at the transition from the B = 1 state to the insulator are also consistent with

these calculations (Huckestein, 1995). This agreement is still a puzzle at this time, especially

as the value of the dynamic scaling exponent z

1 strongly suggests that Coulomb interac-

tions are playing an important role. The evidence for a super-universality of the transitions,

however does have some theoretical support in the form of a set of physically appealing

correspondence rules (Kivelson et al., 1992). Unfortunately, their a priori validity in the

critical regions is still unclear. (Lee and Wang, 1996) In sum, theorists have their work cut

out for them!

IV. CONCLUDING REMARKS, OTHER SYSTEMS

Let us briefly recapitulate our main themes. Zero temperature phase transitions in

quantum systems are fundamentally different from finite temperature transitions in classical

systems in that their thermodynamics and dynamics are inextricably mixed. Nevertheless,

by means of the path integral formulation of quantum mechanics, one can view the statistical

mechanics of a d-dimensional T = 0 quantum system as the statistical mechanics of a d+1

dimensional classical system with a fake temperature which is some measure of zero-point

fluctuations in the quantum system. This allows one to apply ideas and intuition developed

for classical critical phenomena to quantum critical phenomena. In particular this leads to an

understanding of the T = 0 behavior of the quantum system in terms of finite size scaling andto the identification of a T-dependent length scale, L, that governs the crossover between

37


38/62

quantum and classical fluctuations. The identification of QPTs in experiments relies upon

finding scaling behavior with relevant parameters. These are the temperature itself and the

frequency, wavelength and amplitude of various probes. Additional signatures are universal

values of certain dimensionless critical amplitudes such as the special case of resistivities at

critical points in conducting systems in d=2 and, more generally, amplitude ratios.

In this Colloquium we have illustrated these ideas in the context of a single system,

the two dimensional electron gas in the quantum Hall regime. The ideas themselves

are much more widely applicable. Interested readers may wish to delve, for example,

into work on the one dimensional electron gas (Luther and Peschel, 1975; Emery, 1979),

metal insulator transitions in zero magnetic field (Anderson-Mott transitions) (Belitz

and Kirkpatrick, 1994), superconductor-insulator transitions (Wallin, et al., 1994; Cha, et

al., 1991; Chakravarty, et al., 1986; Chakravarty, et al., 1987; Wen and Zee, 1990; Srensen,

et al., 1992; Weichman, 1988; Weichman and Kim, 1989; Batrouni, et al., 1990; Krauth

and Trivedi, 1991; Scalettar, et al., 1991; Krauth et al., 1991; Runge, 1992; Kampf and

Zimanyi, 1993; Batrouni, et al., 1993; Makivic et al., 1993), two-dimensional antiferromag-

nets associated with high temperature superconductivity (Sachdev, 1996; Chakravarty, et

al., 1989; Sachdev and Ye, 1992; Chubukov and Sachdev, 1993; Chubukov, et al., 1994;

Sachdev, et al., 1994; Sachdev, 1994) and magnetic transitions in metals (Hertz, 1976; Mil-

lis, 1993; Sachdev, et al., 1995; Altshuler, et al., 1995). This list is by no means exhaustive

and we are confident that it will continue to expand for some time to come!

ACKNOWLEDGMENTS

It is a pleasure to thank R. N. Bhatt, M. P. A. Fisher, E. H. Fradkin, M. P. Gelfand, S. A.

Kivelson, D. C. Tsui and H. P. Wei for numerous helpful conversations. We are particularly

grateful to K. A. Moler, D. Belitz, S. Nagel, T. Witten, T. Rosenbaum, and U. Zuelicke

for comments on early versions of the manuscript. DS is supported by the NSF, the work

at Indiana is supported by NSF grants DMR-9416906, DMR-9423088 and DOE grant DE-

38


39/62

FG02-90ER45427, and SLS is supported by the NSF through DMR-9632690 and by the A.

P. Sloan Foundation.

39


40/62

APPENDIX A

In this Appendix we briefly outline the derivation (Wallin, et al., 1994) of the expression

for the matrix elements

M {(j+1)}|e h H|{(j)} (A.1)

appearing in Eq. (6). The hamiltonian contains a kinetic energy

T =C

2

j

V2j =EC2

j

i

j

2, (A.2)

where EC (2e)2C , and a potential energy

V EJ cos

j j+1

. (A.3)

For sufficiently small we can make the approximation

ehH e h Te h V. (A.4)

Inserting a complete set of angular momentum eigenstates |{mk} (defined for a singlesite by k|mk = eimkk) yields

M ={m}

{(j+1)}|e h T|{mk}{mk}|e h V|{(j)}. (A.5)

We can now take advantage of the fact that V is diagonal in the angle basis and T is diagonal

in the angular momentum basis to obtain

M ={m}

e2h

EC

km2k eimk[k(j+1)k(j)] e+

hEJ

kcos[k(j+1)k(j)] (A.6)

Because is small, the sum over the {m} is slowly convergent. We may remedy this byusing the Poisson summation formula (Wallin, et al., 1994)

m

e2h

ECm2

eim =

h

EC

n

e h2EC

(+2n)2. (A.7)

This periodic sequence of very narrow gaussians is (up to an irrelevant constant prefactor)

the Villain approximation to

40


41/62

e+ hEC

cos(). (A.8)

Strictly speaking, we should keep infinitesimal. However we may set it equal to the

natural ultraviolet cutoff, the inverse of the Josephson plasma frequency = h/

ECEJ,

without changing the universality class. Substituting this result into Eq.(A.6) yields Eq.(8)

with the same coupling constant K =

EC/EJ in both the space and time directions.

41


42/62

REFERENCES

Aharony, A. and P.C. Hohenberg, 1976, Phys. Rev. B 13, 3081

Ahlers, G., 1980, Rev. Mod. Phys,52

, 489

Altshuler, B.L., L.B. Ioffe, and A.J. Millis, 1995, Phys. Rev. B 52, 5563

Batrouni, G.G., R.T. Scalettar, and G.T. Zimanyi, 1990, Phys. Rev. Lett. 65, 1765

Batrouni, G.G., B. Larson, R.T. Scalettar, J. Tobochnik, and J. Wang, 1993, Phys. Rev. B

48, 9628

Belitz, D. and T.R. Kirkpatrick, 1994, Rev. Mod. Phys. 66, 261

Bradley, R. M. and S. Doniach, Phys. Rev. B 30, 1138

Cha, Min-Chul, M.P.A. Fisher, S.M. Girvin, Mats Wallin, and A. Peter Young, 1991, Phys.

Rev. B 44, 6883

Chaikin, P.M. and T.C. Lubensky, 1995, Principles of Condensed Matter Physics (Cam-

bridge University Press, New York)

Chakraborty, T. and P. Pietilainen, 1995, The Quantum Hall Effects: Integral and

Fractional, 2nd Ed., Springer Series in Solid State Sciences 85, (Springer-Verlag

Berlin/Heidelberg/New York/Tokyo)

Chakravarty, S., G.-L. Ingold, S. Kivelson, and A. Luther, 1986, Phys. Rev. Lett. 56, 2303

Chakravarty, S., S. Kivelson, G.T. Zimanyi, and B.I. Halperin, 1987, Phys. Rev. B 35, 7256

Chakravarty, S., B.I. Halperin, and D.R. Nelson, 1989, Phys. Rev. B 39, 2344

Chow, E., H.P. Wei, S.M. Girvin, and M. Shayegan, 1996, Phys. Rev. Lett. 77, 1143

Chubukov, A.V. and S. Sachdev, 1993, Phys. Rev. Lett. 71, 169

Chubukov, A.V., S. Sachdev, and J. Ye, 1994, Phys.

S. L. Sondhi et al- Continuous Quantum Phase Transitions

Documents

Transcript of S. L. Sondhi et al- Continuous Quantum Phase Transitions